As AI Costs Rise, Value of “Labor Savings” Erodes

When IT/BPO firms began embracing AI and trimming their workforce, they hoped to save a significant amount of money. Initially, they did. But as the months passed, they found that much of those savings were being redirected into AI infrastructure that needed constant maintenance and upgrades.

“I believe some organisations are now spending as much, or even more, on AI than they previously spent on the employees they replaced,” argues Charles Liu, CEO at Cubic Promote, an Australian marketing firm.

Liu’s observations mirror a broader concern emerging across industries. One of the strongest endorsements of that view came from Bryan Catanzaro, VP of Applied Deep Learning at Nvidia. He told Axios: “For my team, the cost of compute is far beyond the costs of the employees.”

I believe some organisations are now spending as much, or even more, on AI than they previously spent on the employees they replaced.

Charles Liu, CEO, Cubic Promote

“So yes, on some workflows AI now costs more than the labor it was supposed to save,” said Miki Furman, CTO of nearshore outsourcing firm Call Force Global. The company operates in several Latin American and Caribbean nations, including Jamaica, Trinidad, Belize, and Colombia.

From Liu to Catanzaro to Furman, the message is much the same: enterprises are not necessarily saving money even after automation.

AI has shifted spending from wages to technology, Furman clarified to Nearshore Americas.

The Initial Days of Automation

When companies first considered automation, they focused mainly on the costs of computing power (GPUs) and AI tokens. What many failed to realize is that maintaining AI infrastructure is expensive, says Kuber Sharma, Senior Director at UiPath, a New York City-based firm that sells automation tools.

“When comparing labor costs, organizations routinely leave that ongoing work off the AI side of the ledger,” Sharma added.

In other words, he says automation is not a one-off job. Deploying AI tools is just the beginning. The real expenses come later. “Models drift, edge cases accumulate, and someone has to own the ongoing engineering work.”

AI models must be regularly updated and improved as they encounter new situations. Consequently, organizations must retain engineers to maintain and fine-tune these systems continuously.

In addition to these ongoing costs, there are hidden costs. “Data preparation, monitoring, cybersecurity, compliance, and integrating AI into existing workflows are some of the costs companies overlook initially,” according to Jacob Ferguson, President of Ferguson Industrial Company.

Overall, Liu argued, “Training and maintaining an AI system can be more expensive and demanding than training an employee.”

As companies deployed more AI tools, their operating expenses rose in tandem, Ferguson added. In the case of BPOs, using more AI tokens increased their spending. At call centers, the higher the call volume, the greater their AI costs.

Interestingly, companies that automated only repetitive tasks neither spent heavily on AI infrastructure nor laid off a significant portion of their workforce, according to Liu.

“I believe enterprises are reconsidering aggressive AI-driven workforce reductions because they are beginning to understand the full cost of maintaining these systems, compared to asking your people to upskill.”

There are no reliable data on how many jobs AI-led automation has eliminated over the past few years. According to layoff data compiled by Challenger, Gray & Christmas and cited by Oxford Economics, AI was named as a reason for roughly 4.5% of publicly reported job cuts in 2025. Many of those job cuts occurred in customer support, software testing, and entry-level administrative positions.

Despite these financial hurdles, AI has undeniably improved the quality of service across outsourcing firms, even if it has failed to improve profit margins.

“The metric shift I’d recommend is to stop comparing AI costs to equivalent headcount costs and start comparing the cost per successful outcome,” Sharma said. “That changes the question from ‘Is AI cheaper than people?’ to ‘Is AI actually delivering the value we deployed it for?”